Benchmarks › GDP.pdf
GDP.pdf
Answer professional questions grounded in long PDF documents
32.6top score · Muse Wonder Full
133models ranked
80open-weights models
All-pass ratehigher is better
Oct 20, 2026last updated
How the test works
1
Task
100 tasks
- What the model doesAnswer professional questions grounded in long PDF documents
- Kinds of tasksDocument reasoning
- In → outText, page images where supported → Answer
2
Run
5 runs per task
- Token budgetNon-reasoning: up to 16,384 output tokens, adjusted to model limits; reasoning: model-specific maximum
- InternetNo
- Temperature0 for non-reasoning; 0.6 for reasoning, unless the model lab recommends otherwise
3
Score
All-pass rate
- Graded byEvery criterion must pass; GPT-5.6 Luna Medium rubric judge
- Better isHigher
Sources: artificialanalysis.ai
GDP.pdf · AA all-pass leaderboardtop 10 of 133
Full leaderboard with filters →| Model | GDP.pdf · AA all-pass | Per task | Price | Speed | Supports | Context |
|---|---|---|---|---|---|---|
32.6 | — | — | 1M | |||
32.2xhigh | $2.31 | $10 / $50 | 61 | 1.05M | ||
32high | $0.32 | $2 / $10 | 50 | 1.05M | ||
30.6max | $4.52 | $4 / $20 | 161 | 1M | ||
28.8high | $1.82 | $4 / $20 | 82 | 1M | ||
28low | $2.37 | $10 / $50 | 57 | 1M | ||
27.8high | $0.81 | $4 / $20 | 85 | 1.05M | ||
26.6max | $1.60 | $1.25 / $4.25 | 128 | 1M | ||
25.8max | $5.46 | $2 / $10 | 137 | 1M | ||
25.2max | $1.04 | $2 / $10 | 128 | 1.05M |
GDP.pdf questions
What is GDP.pdf?›
Answer professional questions grounded in long PDF documents. It has 100 tasks. Created by Surge AI (AA implementation by Artificial Analysis).
Which model is best on GDP.pdf?›
Muse Wonder Full leads GDP.pdf · AA all-pass with 32.6, ahead of GPT-6 Astra (32.2) and GPT-6.1 Sol (32).
What is the best open-weights model on GDP.pdf?›
Kimi K3 with 22, then MiMo v2.6 Pro (19.2).
How is GDP.pdf scored?›
All-pass rate; every criterion must pass; gpt-5.6 luna medium rubric judge; 5 runs per task per task.